DNA Methylation-Based Prognostic Biomarkers of Acute Myeloid Leukemia Patients

737

Original Article Page 1 of 15

DNA methylation-based prognostic biomarkers of acute myeloid leukemia patients

Linjun Hu1#, Yuling Gao2#, Zhan Shi3, Yang Liu1, Junjun Zhao4, Zunqiang Xiao3, Jiayin Lou5, Qiuran Xu6, Xiangmin Tong6

1The Medical College of Qingdao University, Qingdao 266071, China; 2Department of Genetic Laboratory, Shaoxing Women and Children Hospital, Shaoxing 312030, China; 3The Second Clinical Medical College of Zhejiang Chinese Medical University, Hangzhou 310014, China; 4Graduate Department, Bengbu Medical College, Bengbu 233030, China; 5Department of Clinical Laboratory, Da jiang dong Hospital, Hangzhou, 310014, China; 6The Key Laboratory of Tumor Molecular Diagnosis and Individualized Medicine of Zhejiang Province, Zhejiang Provincial People’s Hospital (People’s Hospital of Hangzhou Medical College), Hangzhou 310014, China Contributions: (I) Conception and design: L Hu, Y Gao, Q Xu; (II) Administrative support: X Tong; (III) Provision of study materials or patients: Z Shi, Y Liu, J Lou ; (IV) Collection and assembly of data: J Zhao; (V) Data analysis and interpretation: Z Xiao; (VI) Manuscript writing: All authors; (VII) Final approval of manuscript: All authors. #These authors are co-first authors. Correspondence to: Qiuran Xu, MD, PhD; Xiangmin Tong, MD, PhD. Key Laboratory of Tumor Molecular Diagnosis and Individualized Medicine of Zhejiang Province, Zhejiang Provincial People’s Hospital (People’s Hospital of Hangzhou Medical College), Hangzhou 310014, China. Email: [email protected]; [email protected].

Background: Acute myeloid leukemia (AML) is a heterogeneous clonal disease that prevents normal myeloid differentiation with its common features. Its incidence increases with age and has a poor prognosis. Studies have shown that DNA methylation and abnormal gene expression are closely related to AML. Methods: The methylation array data and mRNA array data are from the Gene Expression Omnibus (GEO) database. Through the GEO data, we identified differential genes from tumors and normal samples. Then we performed Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology (GO) analyses on these differential genes. Protein-protein interaction (PPI) network construction and module analysis were performed to screen the highest-scoring modules. Next, we used SurvExpress software to analyze the genes in the highest-scoring module and selected potential prognostic genes by univariate and multivariate Cox analysis. Finally, the three genes screened by SurvExpress software were analyzed using the methylation analysis site MethSurv to explore AML associated methylation biomarkers. Results: We found three genes that can be used as independent prognostic factors for AML. These three genes are the low expression/methylation genes ATP11A and ITGAM, and the high expression/low methylation gene ZNRF2. Conclusions: In this study, we performed a comprehensive analysis of DNA methylation and gene expression to identify key epigenetic genes in AML.

Keywords: Acute myeloid leukemia (AML); methylation; gene expression; biomarker

Submitted Nov 04, 2019. Accepted for publication Nov 21, 2019. doi: 10.21037/atm.2019.11.122 View this article at: http://dx.doi.org/10.21037/atm.2019.11.122

Introduction and differentially expressed genes (DEGs) in AML remain unclear. This study aims to analyze acute myeloid leukemia (AML); In this study, we performed a comprehensive analysis a group of genetically heterogeneous malignant clonal of DNA methylation and gene expression to identify key diseases that can block normal myeloid differentiation (1), epigenetic genes in AML. The methylation genes and as well as AML which is also the most common type of differential genes of AML patients and normal individuals adult leukemia with the poorest prognosis. were downloaded from the GEO database. After data Only in the United States of 2019, there have been preprocessing, we identified differential genes between 10,920 deaths, and 21,450 new cases of AML diagnosed (2). tumors and normal samples and performed KEGG and At present, induction therapy for AML is based primarily GO analyses on these genes. Protein-protein interaction on cytotoxic drugs that enable complete remission (CR) in (PPI) network construction and module analysis were then 70% of adult patients (3). However, the likelihood of the performed, and the highest-scoring modules were screened. recurrence rate in AML patients is still high, especially in SurvExpress software and analyzed the genes to be assigned older patients with prognosis of having high-risk factors. the highest-scoring module with a P value <0.05 were Yet, the long-term cure rate was lower in the absence of selected to perform survival analysis and risk assessment allogeneic hematopoietic stem cell transplantation (Allo- in the cancer dataset. Finally, MethSurv analyzed the HCT) in adult AML patients with first complete remission three genes screened by SurvExpress software to explore (CR1) (4). methylation biomarkers associated with AML survival. With the emergence of various molecular targeted therapies for AML in the last two years, there are now several new drugs and other clinical trials currently Methods underway; however, there is not enough data to know Microarray data which newly approved drugs should be chosen or in which order or combination they should be used in (2). New We extracted gene expression (GSE114868) and early diagnostic biomarkers and therapeutic targets are methylation (GSE63409) profiling data from the Gene therefore urgently needed to improve the diagnosis and risk Expression Omnibus (GEO) database at the National management of AML in order to reduce the mortality rate Center for Biotechnology Information. of AML. The AML-associated dataset GSE63409 submitted by Epigenetic modifications include a class of heritable Jung N based on the GPL13534 platform was obtained non-genetic changes in gene expression, usually including from the GEO database and included 15 AML samples and DNA methylation, histone modifications, and chromatin 5 normal samples. The AML-associated dataset GSE114868 remodeling (5). In healthy hematopoietic stem cells, submitted by Huang H based on the GPL17586 platform epigenetic processes play a key role in cell differentiation was obtained from the GEO database and included 194 and hematopoiesis (6). Among them, DNA methylation AML samples and 20 normal samples (Figure 1). affects the function of key genes. It is closely related to tumors by silencing tumor suppressor genes and activating Identification of DEGs oncogenes by high/low methylation (7). Abnormal DNA methylation is considered a hallmark of AML and is GEO provides users with a useful tool called GEO2R that considered a powerful epigenetic marker in early diagnosis, can be used to analyze microarray data. GRO2R (https:// prognosis prediction, and treatment decision making (8). www.ncbi.nlm.nih.gov/geo/geo2r/) were used to analyze Abnormal gene expression is closely related to tumor gene expression in GSE114868 and GSE63409. Gene prognosis. Studies have shown that NPM1 (9), FLT3 expression (GSE114868) and methylation (GSE63409) (10,11), C-KIT (12), AML1-ETO (13), RUNX1 (3,14), TP53 data set genes were considered statistically significant as |t| (3,15), CBFB/MYH11 (13,16), TET2 (17), DNMT3A (18), ≥2.0 and adj P<0.01. Then, we identified hypomethylated/ JAK-STAT (19), and CXCR4 (20,21) ,HOXA family (22), upregulated genes via overlapping the hypomethylated and NAT10 (23) gene are associated with prognosis of AML. A overexpression gene lists and identified hypermethylated/ few studies have shown altered DNA methylation in cancer, downregulated genes via overlapping the hypermethylated but the roles of key differentially methylated genes (DMGs) and low-expression gene lists. Finally, Then, we use the

GEO database 1. AML and DNA methylation 2. Homo sapiens 3. Array Datasets incorporating AML and DNA methylation subjects

GEO2R

DMGs and DEGs identification 1. AML and DNA methylation 2. Homo sapiens 3. Array KEGG pathway GO pathway Highest score module enrichment analysis enrichment analysis genetic analysis

SurvExpress

Screen out 3 independent prognostic factors

ROC MethSurv

Establish a prognostic Methylation model biomarkers

Figure 1 Graphical abstract.

Venn diagram network tool to draw the Venn diagram. Then, MCODE plugin in Cytoscape is used to find clusters (http://bioinformatics.psb.ugent.be/webtools/Venn/). with the degree cut-off, haircut on, k-core, node score cutoff, max depth set as 10, 0.2, 2, 0.2 and 100 in PPI network and the module with the highest score would be picked out Functional enrichment analyses for module genes to make survival analyses. The Kyoto Encyclopedia of Genes and Genomes (KEGG) is a database resource for understanding high-level functions Survival analysis and Cox regression, ROC curve and utilities of the biological system from molecular-level information. The Gene Ontology (GO) could be used to SurvExpress is a comprehensive gene expression database perform enrichment analysis. We used DAVID (https:// and web-based tool that uses biomarker gene lists as david.ncifcrf.gov/) to make KEGG pathway analysis and input to provide survival analysis and risk assessment in GO enrichment analysis, including Biological Process, cancer data sets (24). We selected genes in the highest- Cellular Component, and Molecular Function for the scoring module of the hypermethylated low expression hypomethylated/upregulated genes and hypermethylated/ and low methylation high expression overlapping genes, downregulated genes. respectively. The univariate Cox model was used to calculate the association between gene expression levels and patient overall survival (OS). When the P value was <0.05, PPI network construction and module analysis its modular gene was used as an independent prognostic PPI analysis is used to search core genes and gene modules factor for patient survival. AML samples are divided into related to carcinogenesis. In this study, the PPI network two groups: (I) high-risk group; (II) low-risk group. Survival analysis of the hypomethylated/upregulated genes and analysis was performed by Kaplan-Meier survival plots, hypermethylated/downregulated genes were performed log-rank P value, and hazard ratio (HR, 95% confidence using the search tool for the Retrieval of Interacting Genes interval). By comparing the sensitivity and specificity of (STRING) database. The interaction score was set at 0.7. risk-based survival predictions, the accuracy of prognostic

A GSE114868 up B GSE114868 down

6039 273 1304 4730 381 1644 DGI DG2

GSE63409 hypo GSE63409 hyper

Figure 2 Identification of aberrantly methylated differentially expressed genes in gene expression datasets (GSE114868) and gene methylation datasets (GSE63409). (A) Hypomethylation and upregulated genes; (B) hypermethylation and downregulated genes.

A KEGG pathway (top 10) B KEGG pathway (top 10) Thermogenesis ThI and Th2 cell differentiation 5.85 2.88 Epstein-Barr virus infection 5.78 Retrograde endocannabinoid signaling 2.37 Endo 5.32 Oocyte meiosis 2.13 Thl7 cell differentiation 5.20 Transcriptional misregulation in cancer 1.85 Chemokine signaling pathway 4.75 Pathways in cancer 1.84 Phagosome 4.45 Parathy roid hormone synthesis, secretion. 1.82 Human immunodeficiency virus I infection 4.23 Glycine, scrine and thrconine metabolism 1.77 Toxoplasmosis 4.22 Cholinergic synapse 1.73 Human T-cell leukemia virus I infection 4.08 Cushing syndrome 1.70 Human cytomegalovirus infection 3.95 Basal transcription factors 1.63 –log(P-value) 0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 –log(P-value) 0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 Figure 3 Analysis of the top ten pathways of KEGG for aberrant methylation differentially expressed genes in AML. (A) Hypermethylation and downregulated genes pathway names; (B) hypomethylation and upregulated genes pathway names. performance was assessed using ROC curves of patients revealed 273 hypomethylated/upregulated genes and 381 over time, 10, 30, 50, and 70 months, respectively. hypermethylated/downregulated genes (Figure 2A,B).

DNA methylation data in MethSurv KEGG pathway analysis

MethSurv is used to explore methylation biomarkers KEGG analysis was performed with hypomethylated/ associated with the survival of various human cancers (25). high-expression genes and hypermethylated/low- MethSurv is freely available at https://biit.cs.ut.ee/methsurv. expression genes, and the first ten pathways were selected. Through the MethSurv website, we will analyze the DNA Hypermethylated/low-expression genes were enriched methylation analysis of the selected AML-related genes in in Th1 and Th2 cell differentiation, Epstein-Barr the TCGA database. virus infection, Endocytosis, Th17 cell differentiation, Chemokine signaling pathway, Phagosome, Human Results immunodeficiency virus 1 infection, Toxoplasmosis, Human T-cell leukemia virus 1 infection, Human cytomegalovirus Identification of DEGs and DMGs in AML infection (Figure 3A). The hypomethylated/high-expression After obtaining DEGs and DMGs, 6,039 upregulated genes genes were significantly enriched in the thermogenesis, and 4,730 low-expression genes in GSE114868 were found retrograde endocannabinoid signaling, oocyte meiosis, from the gene expression microarray analysis. There were transcriptional misregulation in cancer, pathways in cancer, 1,644 hypermethylated genes and 1,304 hypomethylated parathyroid hormone synthesis, secretion and action, genes in GSE63409 identified by the gene methylation glycine, serine and threonine metabolism, cholinergic microarray analysis. Further analysis of overlapping genes synapse, Cushing syndrome, basal transcription factors

© Annals of Translational Medicine. All rights reserved. Ann Transl Med 2019;7(23):737 | http://dx.doi.org/10.21037/atm.2019.11.122 Annals of Translational Medicine, Vol 7, No 23 December 2019 Page 5 of 15 4.50 3.97 6.7 4.06 3.42 3.30 5.35 3.29 4.82 3.08 4.73 2.72 4.6 2.63 2.97 2.57 2.57 4.43 4.41 4.41 4.38 2.78 2.42 2.38 2.32 2.62 2.42 2.35

GO Biological Process (TOP 10) GO Biological Process GO Cellular Component (TOP 10) –log(P-value) 0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00 4.50 GO Molecular Function (TOP 10) –log(P-value) 0.00 0.05 1.00 1.50 2.00 2.50 3.00 3.50 4.00 4.50 5.00 –log(P-value) 0 1 2 3 4 5 6 7 8 RNA binding (GO:000.3723) Tertiary granule (GO:0070820) Tertiary Helicase activity (GO:0004386) Specific granule (GO:0042581) Early endosome (GO:0005769) Phosphory lation ( GO:0016310) Actin cytoskeleton (GO:0015629) Phagocytic vesicle (GO:0045335) Protein kinase binding (GO:0019901) Protein Damaged DNA binding(GO:0003684) Protein phosphorylation (GO:0006468) Protein Tertiary granule membrane (GO:0070821) Tertiary Phagocytic vesicle membrane (GO:0030670) Regulation of apoptotic process (GO:0042981) Regulation of apoptotic process Transcription coactivator activity (GO:0003713) Transcription Phosphatidylethanol amine binding(GO:0008429) Hyaluronoglucosaminidase activity (GO:0004415) Hyaluronoglucosaminidase Cytokine-mediated signaling pathway (GO:0019221) Regulation of interleukin-12 production (GO:0032655) Regulation of interleukin-12 production N-acety Neuraminate metabolic process (GO:0006054) N-acety Neuraminate metabolic process Positive regulation of leukocyte degranulation (GO:0043302) Positive regulation Integral component of endoplasmic reticulum membrane (GO:000176) Integral component of endoplasmic reticulum Positive regulation of myeloid leukocyte mediated immunity (GO:0002888 Positive regulation S-adenosylmethionine-dependent methyltransferase activity (GO:0008757) Divalent inorganic cation transmembrane transporter activity (GO:0072509) Divalent inorganic Positive regulation of NF-kappaB transcription factor activity (GO:0051092) Positive regulation Proton-transporting ATP synthase acti vity, rotational mechanism (GO:0046933) rotational synthase acti vity, ATP Proton-transporting F B D Integral component of lumenal side of endoplasmic reticulum membrane (GO:0071556) Integral component of lumenal side endoplasmic reticulum 5.65 3.70 3.47 3.42 2.99 3.40 3.30 3.27 3.27 2.57 2.50 2.44 2.81 2.43 2.65 2.62 2.17 3.76 3.75 2.04 2.01 1.95 3.39 3.28 3.26 1.88 3.90 2.85 2.77 2.70

GO Cellular component (TOP 10) GO Biological Process (TOP 10) GO Biological Process GO Molecular Function (TOP 10) –log(P-value) 0.00 1.00 2.00 3.00 4.00 5.00 6.00 –log(P-value) 0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00 Nuclear body (GO:001660) –log(P-value) 0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 Nuclear speck (GO:0016607) Actin binding (GO:0003779) Cytoplasmic vesicle (GO:0031410) Mitochondrial matrix (GO:0005759) PDZ domain binding (GO:0030165) Glial cell development (GO:0021782) Actin filament binding (GO:0051015 Protein kinase activity (G0: 0004672) Protein Protein kinase binding (GO:001990n) Protein Toll-like receptor binding (GO:0035325) receptor Toll-like Cytoplasmic esicle membrane (GO:0030659) L-serine metabolic process (GO:0006563) L-serine metabolic process Protein phosphatase binding (GO:0019903) Protein Protein ty rosine kinase activity (GO:0004713) ty rosine Protein CulA-RING E3 ubiquitin ligase complex(GO:003146) Cytoplasmic ribonucleoprotein granule(GO:0036-64) Cytoplasmic ribonucleoprotein Negative regulation of gene expression (GO:0010629) of gene expression Negative regulation Regulat ion of histone H3-K9 methylation (GO:0051570) Positive regulation of histone methy lation (GO:0031062) Positive regulation Protein kinase A catalytic subunit binding (GO:0034236) Protein Regulation of Ras protein signal transduction (GO:0046578) Regulation of Ras protein Negative regulation of gene expression, epigenetic (GO:0045814) of gene expression, Negative regulation Analysis of the top ten pathways GO for aberrant methylation differentially expressed genes in AML. (A,B,C) Hypomethylation and upregulated pathway Mitochondrial proton-transporting ATP synthase complex(GO:0005753) ATP Mitochondrial proton-transporting

Regulation of methylation-dependent chromatin silencing (GO:0090308) Regulation of methylation-dependent chromatin Non-membrane spanning protein tyrosine kinase activity (GO:0004715) tyrosine Non-membrane spanning protein Intrinsic component of the cytoplasmic side plasma membrane (GO:0031235) E A Negative regulation of small GTPase mediated signal transduction (GO:0051058) Negative regulation C Regulation of intracellular steroid hormone receptor signaling pathway (GO:0033143) hormone receptor Regulation of intracellular steroid Mitochondrial proton-transporting ATP synthase complex, coupling factor F(o) (GO:0000276) ATP Mitochondrial proton-transporting Figure 4 names; (D,E,F) hypermethylation and downregulated genes pathway names.

(Figure 3B). tyrosine kinase activity, PDZ domain binding, protein kinase binding, protein kinase A catalytic subunit binding, protein phosphatase binding (Figure 4E). In the CC, the GO term analysis hypermethylated/low-expression genes were enriched GO analysis was performed with hypomethylated/high- in the phagocytic vesicle, phagocytic vesicle membrane, expression genes and hypermethylated/low-expression tertiary granule membrane, is an integral component of genes, and the first ten pathways were selected. Regarding endoplasmic reticulum membrane, specific granule, actin the biological processes (BP), the hypomethylated/ cytoskeleton, which is integral component of lumenal side upregulated genes were significantly enriched in regulation of endoplasmic reticulum membrane, early endosome, of methylation-dependent chromatin silencing, positive tertiary granule, Golgi membrane (Figure 4F). regulation of histone methylation, regulation of histone H3-K9 methylation, negative regulation of gene expression, PPI network construction and module choice epigenetic, L-serine metabolic process, regulation of intracellular steroid hormone receptor signaling pathway, PPI network analysis was performed through the STRING negative regulation of small GTPase mediated signal database and Cytoscape software. We obtained a PPI transduction, regulation of Ras protein signal transduction, network map of hypomethylation/upregulation genes and negative regulation of gene expression, glial cell hypermethylation/downregulation genes. The PPI network development (Figure 4A). In the molecular function (MF), map of the hypomethylated/upregulated gene has 256 nodes the hypomethylated/high-expression genes were enriched in and 176 edges (Figure 5), and the PPI network map of the helicase activity, proton-transporting ATP synthase activity, hypermethylated/downregulated gene has 373 nodes and rotational mechanism, protein kinase binding, divalent 309 edges (Figure 6). The above two PPI network diagrams inorganic cation transmembrane transporter activity, were respectively imported into the Cytoscape software to transcription coactivator activity, phosphatidylethanolamine construct a module through the MCODE plug-in, in which binding, hyaluronoglucosaminidase activity, damaged the hypomethylation/upregulation gene was constructed DNA binding, S-adenosylmethionine-dependent into 9 modules, and the hypermethylation/downregulation methyltransferase activity, RNA binding (Figure 4B). In gene was constructed into 12 modules. We selected the the cellular component (CC), the analysis revealed that highest scoring modules (Figure 7A,B). enrichment mainly occurred at intrinsic component of the cytoplasmic side of the plasma membrane, nuclear speck, Survival analysis of genes mitochondrial proton-transporting ATP synthase complex, mitochondrial matrix, mitochondrial proton-transporting To assess whether the identified prognostic markers are ATP synthase complex, coupling factor F(o), cytoplasmic valuable for predicting patient survival, we focused on the vesicle membrane, cytoplasmic ribonucleoprotein granule, genes in the highest-scoring modules. We used SurvExpress cytoplasmic vesicle, nuclear body, Cul4A-RING E3 software to analyze hypomethylated/upregulated, and ubiquitin ligase complex (Figure 4C). hypermethylated/downregulated genes for univariate The BP enriched by the hypermethylated/low- and multivariate Cox regression analysis. AML samples expression genes included cytokine-mediated signaling were divided into two groups: (I) high-risk group; (II) pathway, protein phosphorylation, positive regulation of low-risk group. First, we performed the univariate Cox myeloid leukocyte mediated immunity, phosphorylation, regression analysis of the highest-scoring module gene. The N-acetylneuraminate metabolic process, positive regulation hypermethylation/downregulation gene module 1 has 20 of NF-kappaB transcription factor activity, positive genes, and the hypomethylation/upregulation gene module regulation of leukocyte degranulation, regulation of 1 contains 7 genes (Table 1). We screened for genes with interleukin-12 production, regulation of apoptotic process, P values <0.05 (ATP11A, ITGAM, ZNRF2). These three interferon-gamma-mediated signaling pathway (Figure 4D). genes were then subjected to multivariate Cox regression In the MF, the hypomethylated/high-expression genes analysis and found to have P values <0.05 (Table 2). The were enriched in protein kinase activity, actin filament three genes were independent prognostic factors, and binding, non-membrane spanning protein tyrosine kinase then we use three genes to establish a prognostic model activity, actin-binding, Toll-like receptor binding, protein (Figure 8A,B). The ROC curves assessed the accuracy of the

Figure 5 PPI network of hypomethylation/upregulation genes.

Figure 6 PPI network of hypermethylation/downregulation genes.

Table 1 The highest scoring module gene Hypermethylation/downregulation module 1 genes

PTGDR

TRPM2

OLR1

ATP11A

GPR25

F2RL1 ZNRF2 CXCR5

CD177 ASB13

TRIM71 ATP6V0C

EDNRB

RNF14 TAS1R2

AVP

CDC16 GRK5 FBXO41 INSL3

ITGAM ANAPC10 CXCL16 B CXCL5

OPRM1

TAS1R2 GPHB5 CXCL5 CXCL16 PTAFR

CXCR5 Hypomethylation/upregulation module 1 genes

OPRM1 FBXO41

CDC16

ASB13 GRK5 ANAPC10 ITGAM ATP Y C ATP F2RL1 RNF14 EDNRB PTAFR CD177 TRP1/2 ZNRF2 AVP

INSL3 TRIM71 ATP11A OLR1 PTGDR GPR25 GPHB5 Table 2 Three survival-related genes based on multivariate Cox A regression analysis Genes HR 95% CI P value

ATP11A 0.599 0.441–0.813 0.001

ITGAM 1.305 1.144–1.488 <0.001 The highest scoring module. (A) Hypermethylation/downregulation gene module 1; (B) hypomethylation/upregulation 1. ZNRF2 0.579 0.408–0.822 0.002 Figure 7

Survival curve A B Color Key & Histogram 70 Low risk (n=84) 40

High risk (n=84) Count 1.0 10 3 4 5 6 7 Log-Rank equal curves P=0.0002179 Value 2-Low risk :84 Risk 0.8 Hazard ratio =2.07 (95% CI: 1.39–3.07) 1-High risk :84

ZNRF2 0.6

0.4

Survival rate ATP11A

0.2

0.0 ITGAM

0 20 40 60 80 84 55 36 25 21 14 8 4 2 84 41 22 10 6 3 0 0 0 Time (months)

C D Gene expression by risk group Survival ROC

1.0 P=7.43e-04 P=1.58e-11 P=2.66e-09

0.8 8

0.6 6

0.4 4 True positives True Time=10, AUC=0.671 Gene expression level Gene expression Time=30, AUC=0.722 0.2 2 Time=50, AUC=0.695

Time=70, AUC=0.754 ATP11A ITGAM ZNRF2 0.0

0.0 0.2 0.4 0.6 0.8 1.0 False positives

Figure 8 Multivariate Cox Regression analysis of three genes. (A) Kaplan-Meier survival curves for overall survival outcomes according to the risk cutoff point; (B) the heatmap of three independent AML-related prognostic genes. 2 genes were Hypermethylation/downregulation genes, and 1 gene were Hypomethylation/upregulation genes; (C) time-dependent ROC curves analysis for survival prediction by the three key genes; (D) Box plot of three key genes expression by risk groups.

prognostic model at 10, 30, 50, and 70 months (Figure 8C). DNA methylation data in MethSurv These three genes were used to evaluate the prognosis of AML cytogenetic risk typing and cell morphological MethSurv is used to explore methylation biomarkers typing and found that these three genes differed in the associated with the survival of various human cancers. intermediate/Normal and M2 prognostic analysis (Table 3). MethSurv analyzed DNA methylation in TCGA. We Finally, we show the expression levels and risk assessment of performed methylation analysis of three genes (ATP11A, three key genes (Figure 8D). ITGAM, ZNRF2) that can be used as independent

Table 3 Cytogenetic risk typing and morphological cell typing of these three genes based on multivariate Cox regression analysis in AML Class HR 95% CI P value

CYTO-RISK-Favorable 2.750 0.570–13.300 0.190

CYTO-RISK-Intermediate/normal 1.750 1.050–2.900 0.029

CYTO-RISK-Poor 1.360 0.570–3.250 0.475

MORPHO-CODE-M0 undifferentiated 1.090 0.300–3.910 0.895

MORPHO-CODE-M1 1.750 0.740–4.100 0.195

MORPHO-CODE-M2 2.850 1.220–6.660 0.010

MORPHO-CODE-M3 3.990 0.410–38.940 0.198

MORPHO-CODE-M4 1.530 0.680–2.440 0.295

MORPHO-CODE-M5 0.880 0.240–3.200 0.849 AML, acute myeloid leukemia.

Table 4 The significant prognostic value of CpG in three key genes Gene-CpG HR LR test P value

ATP11A-Body-Open_Sea-cg00990020 2.207 <0.001

ATP11A-Body-N_Shelf-cg04656015 2.273 <0.001

ATP11A-Body-N_Shelf-cg24199463 2.224 0.001

ATP11A-Body-S_Shelf-cg04229372 2.130 <0.001

ATP11A-Body-Open_Sea-cg00717219 1.837 0.001

ATP11A-Body-Open_Sea-cg25347801 1.826 0.002

ATP11A-Body-Open_Sea-cg21649349 1.975 0.004

ATP11A-Body-N_Shelf-cg03353885 1.881 0.002

ATP11A-Body-S_Shelf-cg09804528 1.866 0.002

ATP11A-Body-Open_Sea-cg17218041 1.884 0.008

ITGAM-Body-Island-cg05625471 1.558 0.036

ITGAM-Body-Island-cg02256631 1.523 0.045

ZNRF2-Body-Open_Sea-cg07510230 0.581 0.014

ZNRF2-Body-Open_Sea-cg07568841 0.655 0.023

ZNRF2-TSS1500-N_Shore-cg21557180 1.556 0.046

prognostic factors for AML in SurvExpress software. sites with a P value of <0.05, which is statistically significant In the MethSurv software, we found that the P value of (Table 4). We found that the difference in DNA methylation 73 CpG sites in the hypermethylated/down-regulated between cg00990020 of ATP11A, cg05625471 of ITGAM, and ATP11A in AML was <0.05, which we considered cg07510230 of ZNRF2 was most pronounced (Figure 9A,B,C). statistically significant. The top ten sites were shown in Table 4. Similarly, the hypermethylation/downregulation gene Discussion ITGAM has two CpG sites with a P value of <0.05, and the hypomethylation/upregulation gene ZNRF2 has three CpG AML is the most common acute leukemia in adults and is

A ATP11A B ITGAM

Ethnicity Ethnicity Ethnicity Ethnicity Race [Not available] Race [Not available] Age 0.8 Hispanic or latino Age 0.8 Hispanic or latino Event Event Not hispanic or latino Not hispanic or latino 0.6 0.6 Race Race cg10371560 [Not available] 0.4 [Not available] 0.4 Asian Asian Black or African American Black or African American 0.2 White cg00764981 0.2 White Age Age (18,45] (18,45] (45,58] (45,58] (58,67] cg03458229 (58,67] (67,88] (67,88]

Event Event Alive cg09743950 Alive Dead Dead Relation_to_UCSC_CpG_Island Relation_to_UCSC_CpG_Island Island cg02846316 Island N_Shelf N_Shelf N_Shore N_Shore Open_Sea Open_Sea S_Shelf cg08050115 S_Shore S_Shore UCSC_RefGene_Group UCSC_RefGene_Group 3'UTR 1stExon cg15337006 Body 3'UTR TSS1500 3'UTR Body TSS200 Body TSS1500 cg00833777

cg22490695

cg02256631

cg05625471

cg09326409

cg15817542

C ZNRF2

Ethnicity Ethnicity Race [Not available] Age 0.8 Hispanic or latino Event Not hispanic or latino 0.6 cg05357932 Race [Not available] 0.4 Asian cg09470758 Black or African American 0.2 White cg26176103 Age (18,45] (45,58] cg08586494 (58,67] (67,88]

cg27662633 Event Alive Dead cg10289912 Relation_to_UCSC_CpG_Island Island cg04522931 N_Shore Open-Sea cg16149129 S_Shore UCSC_RefGene_Group 1stExon cg17460047 1stExon;5'UTR 3'UTR cg07003744 Body TSS1500 TSS200 cg04944410

cg07126792

cg23640092

cg08549335

cg21557180

cg07568841

cg07510230

cg16238026

Figure 9 DNA methylation of three key genes in MethSurv. (A) ATP11A; (B) ITGAM; (C) ZNRF2.

© Annals of Translational Medicine. All rights reserved. Ann Transl Med 2019;7(23):737 | http://dx.doi.org/10.21037/atm.2019.11.122 Annals of Translational Medicine, Vol 7, No 23 December 2019 Page 13 of 15 a highly heterogeneous and fatal (1). DNA methylation is cell proliferation (33). ZNRF2 also plays a crucial role in the most studied epigenetic modification and is essential tumorigenesis. For example, ZNRF2 enhances the mTor in the promotion of many important BP (7). Abnormal and its downstream targets CyclinD1 and CDK in NSCLC DNA methylation can lead to a variety of pathological cells, and the negative correlation between ZNRF2 and conditions, including carcinogenesis. (8) Certain DNA miR-100 in osteosarcoma specimens, low miR-100 is methylation engages in the initial stages of carcinogenesis, associated with poor prognosis in OS patients (34,35). such as RASSF1A in ovarian cancer (26). In addition, To date, the role of the ZNRF2 gene in AML and how it DNA methylation is stable over some time and can be regulates AML through aberrant methylation is unclear. detected non-invasively in the blood (27). Therefore, DNA These three genes may be good indicators for assessing the methylation has the exciting potential to become an early prognosis of AML. We used these three genes to set up an diagnostic biomarker for cancer. Multiple studies have independent prognostic model with high accuracy, which shown that gene expression abnormalities are closely related can be used to assess the prognosis of patients with AML to AML (10). Identification of novel biomarkers will aid in and as a good target for AML treatment. early diagnosis and improved prognosis. In this study, we This study had several limitations: (I) the small number performed a comprehensive analysis of DNA methylation of cases evaluated; (II) the results of the study have not been and gene expression to identify key epigenetic genes in validated on clinical samples. AML. In summary, our research identified many aberrantly This study is the first to report bioinformatics studies expressed genes and pathways that can be regulated by and its association between gene methylation of AML aberrant methylation in AML through a comprehensive and the corresponding mRNA expression. Through analysis of gene expression and methylation microarrays. our research, we found three genes that can be used as We identified some new markers and pathways through independent prognostic factors for AML. These three multi-database analysis that could be an accurate diagnosis genes are the low expression/methylation genes ATP11A and treatment for AML; however, determining the role of and ITGAM, and the high expression/low methylation these identified genes in the AML process requires more gene ZNRF2. ATP11A is an adenosine triphosphate research. binding cassette (ABC) transporter homolog gene and belongs to an extended family of ABC transporters that Acknowledgments confer multidrug resistance to cancer cells. For example, in lymphocytic leukemia, cancer cells are resistant by Funding: This study was supported by grants from the increasing ATP11A expression (28). In previous studies, National Natural Science Foundation of China (81874049, it was found that the expression level of ATP11A gene in 81602179, 81570198); the National Science and Technology colorectal cancer tumor tissues was significantly higher Major Project for New Drug (No. 2017ZX301033); the than that in corresponding normal tissues, and it was Co-construction of Provincial and Department Project important for the prognosis evaluation of colorectal (WKJ-ZJ-1919); the Zhejiang Provincial Natural Science cancer (29). Studies have shown that the ATP11A gene is a Foundation of China (LY19H160036). This study was methylation biomarker for prostate cancer and is expressed supported by the Key Laboratory of Tumor Molecular in patients with metastatic and lethal PCA (30). ITGAM Diagnosis and Individualized Medicine of Zhejiang is a major non-human leukocyte antigen associated with Province, Zhejiang Provincial People’s Hospital (People’s the pathogenesis of autoimmune diseases such as systemic Hospital of Hangzhou Medical College). Hangzhou Health lupus erythematosus (SLE) and IgA nephropathy (31). Science and Technology Project (No. Y201869156). Recent reports that SNP rs4597342 in ITGAM 3'UTR affect miR-21 binding may be considered a risk factor for Footnote psoriasis development (32). However, the above two genes have not been reported in AML. ZNRF2 is a ubiquitin Conflicts of Interest: The authors have no conflicts of interest ligase of the RING superfamily. It has been shown that to declare. membrane-associated E3 ubiquitin ligase ZNRF2 is involved in mTor activation and regulation through protein Ethical Statement: The authors are accountable for all aspects interactions, and ZNRF2 depletion reduces cell size and of the work, and the questions related to the accuracy

© Annals of Translational Medicine. All rights reserved. Ann Transl Med 2019;7(23):737 | http://dx.doi.org/10.21037/atm.2019.11.122 Page 14 of 15 Hu et al. Prognostic markers of AML methylation genes or integrity of any part of the work are appropriately therapy for AML expressing somatic or germline mutation investigated and resolved. in RUNX1. Blood 2019;134:59-73. 15. Hunter AM, Sallman DA. Current status and new treatment approaches in TP53 mutated AML. Best Pract References Res Clin Haematol 2019;32:134-44. 1. Nagler E, Xavier MF, Frey N. Updates in 16. Monma F, Nishii K, Shiga J, et al. Detection of the CBFB/ immunotherapy for acute myeloid leukemia. Transl MYH11 fusion gene in de novo acute myeloid leukemia Cancer Res 2017;6:86-92. (AML): a single-institution study of 224 Japanese AML 2. Lai C, Doucette K, Norsworthy K. Recent drug approvals patients. Leuk Res 2007;31:471-6. for acute myeloid leukemia. J Hematol Oncol 2019;12:100. 17. Mohamed AM, Balsat M, Koering C, et al. TET2 exon 3. Xu P, Wang M, Jiang Y, et al. The association between 2 skipping is an independent favorable prognostic factor expression of hypoxia inducible factor-1α and multi-drug for cytogenetically normal acute myelogenous leukemia resistance of acute myeloid leukemia. Transl Cancer Res (AML): TET2 exon 2 skipping in AML. Leuk Res 2017;6:198-205. 2017;56:21-8. 4. Vasu S, Kohlschmidt J, Mrózek K, et al. Ten-year outcome 18. Spencer DH, Russler-Germain DA, Ketkar S, et al. CpG of patients with acute myeloid leukemia not treated with Island Hypermethylation Mediated by DNMT3A Is a allogeneic transplantation in first complete remission. Consequence of AML Progression. Cell 2017;168:801- Blood Adv 2018;2:1645-50. 816.e13. 5. Koutsi A, Vervesou EC. Diagnostic molecular techniques 19. Faderl S, Ferrajoli A, Harris D, et al. WP-1034, a novel in haematology: recent advances. Ann Transl Med JAK-STAT inhibitor, with proapoptotic and antileukemic 2018;6:242. activity in acute myeloid leukemia (AML). Anticancer Res 6. Kramer A, Challen GA. The epigenetic basis of 2005;25:1841-50. hematopoietic stem cell aging. Semin Hematol 20. Du W, Lu C, Zhu X, et al. Prognostic significance of 2017;54:19-24. CXCR4 expression in acute myeloid leukemia. Cancer 7. Kulis M, Esteller M. DNA methylation and cancer. Adv Med 2019;8:6595-603. Genet 2010;70:27-56. 21. Shah MV, Barochia A, Loughran TP. Impact of genetic 8. Yang X, Wong MPM, Ng RK. Aberrant DNA Methylation targets on cancer therapy in acute myelogenous leukemia. in Acute Myeloid Leukemia and Its Clinical Implications. Adv Exp Med Biol 2013;779:405-37. Int J Mol Sci 2019. doi: 10.3390/ijms20184576. 22. Chen SL, Qin ZY, Hu F, et al. The Role of the HOXA 9. Heath EM, Chan SM, Minden MD, et al. Biological Gene Family in Acute Myeloid Leukemia. Genes 2019. and clinical consequences of NPM1 mutations in AML. doi: 10.3390/genes10080621. Leukemia 2017;31:798-807. 23. Liang P, Hu R, Liu Z, et al. NAT10 upregulation indicates 10. Ferrara F, Palmieri S, Leoni F. Clinically useful prognostic a poor prognosis in acute myeloid leukemia. Curr Probl factors in acute myeloid leukemia. Crit Rev Oncol Cancer 2019. [Epub ahead of print]. Hematol 2008;66:181-93. 24. Aguirre-Gamboa R, Gomez-Rueda H, Martínez-Ledesma 11. Hillert LK, Bettermann-Bethge K, Nimmagadda SC, et al. E, et al. SurvExpress: an online biomarker validation tool Targeting RIPK1 in AML cells carrying FLT3-ITD. Int J and database for cancer gene expression data using survival Cancer 2019;145:1558-69. analysis. PLoS One 2013;8:e74250. 12. Chen PY, Chen YT, Gao WY, et al. Nobiletin Down- 25. Modhukur V, Iljasenko T, Metsalu T, et al. MethSurv: a Regulates c-KIT Gene Expression and Exerts web tool to perform multivariable survival analysis using Antileukemic Effects on Human Acute Myeloid Leukemia DNA methylation data. Epigenomics 2018;10:277-88. Cells. J Agric Food Chem 2018;66:13423-34. 26. Si JG, Su YY, Han YH, et al. Role of RASSF1A promoter 13. Singh AA, Mandoli A, Prange KH, et al. AML associated methylation in the pathogenesis of ovarian cancer: a meta- oncofusion proteins PML-RARA, AML1-ETO and analysis. Genet Test Mol Biomarkers 2014;18:394-402. CBFB-MYH11 target RUNX/ETS-factor binding sites 27. Laird PW. The power and the promise of DNA to modulate H3ac levels and drive leukemogenesis. methylation markers. Nat Rev Cancer 2003;3:253-66. Oncotarget 2017;8:12855-65. 28. Zhang B, Groffen J, Heisterkamp N. Resistance to 14. Mill CP, Fiskus W, DiNardo CD, et al. RUNX1-targeted farnesyltransferase inhibitors in Bcr/Abl-positive

lymphoblastic leukemia by increased expression of 32. Hruska P, Kuruczova D, Vasku V, et al. MiR-21 binding a novel ABC transporter homolog ATP11a. Blood site SNP within ITGAM associated with psoriasis 2005;106:1355-61. susceptibility in women. Plos One 2019;14:e0218323. 29. Miyoshi N, Ishii H, Mimori K, et al. ATP11A is a 33. Hoxhaj G, Caddye E, Najafov A, et al. The E3 ubiquitin novel predictive marker for metachronous metastasis of ligase ZNRF2 is a substrate of mTORC1 and regulates colorectal cancer. Oncol Rep 2010;23:505-10. its activation by amino acids. Elife 2016. doi: 10.7554/ 30. Zhao S, Geybels MS, Leonardson A, et al. Epigenome- eLife.12278. Wide Tumor DNA Methylation Profiling Identifies Novel 34. Xiao Q, Yang Y, An Q, et al. MicroRNA-100 suppresses Prognostic Biomarkers of Metastatic-Lethal Progression human osteosarcoma cell proliferation and chemo- in Men Diagnosed with Clinically Localized Prostate resistance via ZNRF2. Oncotarget 2017;8:34678-86. Cancer. Clin Cancer Res 2017;23:311-9. 35. Zhang XF, Guo ZQ, Zhao CC, et al. The role of ZNRF2 31. Fan Y, Li LH, Pan HF, et al. Association of ITGAM in the growth of non-small cell lung cancer. Eur Rev Med polymorphism with systemic lupus erythematosus: a meta- Pharmacol Sci 2016;20:4011-7. analysis. J Eur Acad Dermatol Venereol 2011;25:271-5.

Cite this article as: Hu L, Gao Y, Shi Z, Liu Y, Zhao J, Xiao Z, Lou J, Xu Q, Tong X. DNA methylation-based prognostic biomarkers of acute myeloid leukemia patients. Ann Transl Med 2019;7(23):737. doi: 10.21037/atm.2019.11.122